Optimization-based calibration of a static timing analyzer to path delay measurements

Values for the entire set of technology parameters in a static timing analyzer are determined by a single run of a nonlinear constrained-optimization package, so as to minimize the sum of squares of errors on a set of benchmark paths, relative to SPICE, while simultaneously requiring each error to be conservative. The new approach yields two benefits. First, the error is significantly reduced: Compared to the conventional approach, average error is reduced by 26% on a representative set of paths. Second, the conception and implementation of model improvements in the static timing analyzer is greatly simplified.